Trust Adaptation Leads to Lower Control Effort in Shared Control of Crane Automation

We present a shared-control framework predicated on a measure of trust in the operator, that is calculated automatically based on the quality of the interactions between a human and autonomous system. This measure of trust is built upon a control-theoretic foundation that rewards stable operation of the system to give more trusted users additional control authority. The level of control authority is used to modify the human input, and as a result, we observe a minimization of the required effort of the controller. We validate this work within a planar crane environment with a receding horizon controller to assist with the regulation of the system dynamics. The human defines the reference trajectory for the controller. In an experimental study users navigate a suspended payload through a set of maze configurations. We find that adaptation of the trust metric over time provides the benefit of substantially ( $p <; 0.01$) improving the automated system's ability to modulate the user's input, resulting in stable reference trajectories that require less effort to track. In effect, the human and automation spend less time fighting each other during task execution, suggesting that the automated system and user each have a better understanding of the other's ability.